H2OAutoML.fit() fails with "Wrong FS" exception when keepBinaryModels=True on Windows

Question

Training works fine with keepBinaryModels=False, but when set True I get:

Traceback (most recent call last):
  File "C:\spark\spark-3.2.4-bin-hadoop2.7\python\pyspark\sql\utils.py", line 117, in deco
    raise converted from None
pyspark.sql.utils.IllegalArgumentException: Wrong FS: file://C:\Users\v-kefree\AppData\Local\Temp\spark-c8c76f0e-1caf-4658-9543-3935c83b4d7e\spark-9a44f7b4-3987-4367-8315-3a1ddf231b2d\StackedEnsemble_AllModels_1_AutoML_1_20230429_101154.bin, expected: file:///

Looks like windows wants "file:///" while sparkling-water is using "file://". I have added the winutils.exe for hadoop 2.7 in my hadoop bin. I've created the hive-site.xml and followed all the instructions from here: https://docs.h2o.ai/sparkling-water/3.2/latest-stable/doc/deployment/run_on_windows.html#

This is sparkling-water 3.40.0.2-1-3.2 running on spark-3.2.4-bin-hadoop2.7. I believe I must set it to true to get the model so I can save it off and use it in a regular h2o (non sparkling) environment, so I'm stuck.